Design and Implementation of a Textureless Object Recognition Algorithm Based on Image Edge Features

• Main Authors: Chao-Chun Yu, 游詔鈞
• Other Authors: Chi-Yi Tsai
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/89433183125561391764

碩士 === 淡江大學 === 電機工程學系碩士班 === 104 === Object recognition is one of the important tasks in the field of computer vision, and most of the conventional methods use texture information of objects to produce feature descriptors for object recognition process. However, in many practical applications, the object to be recognized may not have enough texture information for extracting feature descriptors, greatly increasing the difficulty of the object recognition task. This problem generally is referred to as textureless object recognition. Therefore, how to solve the problem of textureless object recognition is an important issue in practical applications. In this thesis, a textureless object recognition method is proposed based on the existing Line2D algorithm. The proposed method employs an edge-based template matching method to detect and identify a wide variety of textureless objects. Given a reference template image of an object-of-interest (OOI), a template image database containing various postures of the OOI was firstly created by applying affine transformation with different rotating and scaling settings to the reference template image. Next, the edge-based template matching process is performed to detect and recognize the OOI by searching matches between the template image database and the input image. Finally, the position and angle posture of the OOI can be determined by the best match having the highest similarity measure. Experimental results show that the proposed method not only can efficiently recognize the type, position, and angle information of various textureless objects in the image, but also can identify up-down relationship between the recognized objects. In addition, the proposed method achieves real-time performance at least 23 frames per second (fps) in processing 640x480 images. In future work, the proposed object recognition algorithm will be integrated into a robot manipulator system to accomplish random bin-picking function for manipulating textureless objects.